$60–70/hr · Mercor · Part time, 40 hours a week
A productivity software expert who grades and designs tasks for AI training on office tools, evaluating whether AI outputs work under actual professional use.
What you would do
- Author realistic work samples based on your own professional documents, specifying starting materials, constraints, and the correct result that AI attempts must match
- Review AI-generated spreadsheets to verify formulas calculate accurately, handle edge cases, and remain robust when edited or moved to different files
- Evaluate document layouts, ensuring styling persists across file formats and collaborators don't lose structure when making changes
- Analyze slide decks for logical flow, chart correctness, and whether the visual argument holds up in high-stakes presentations
- Document technical failures clearly for research teams, explaining both what broke and why the failure matters in production environments
Who they want
- Extensive hands-on experience in office and productivity tools-documents, spreadsheets, presentations, or email platforms-used daily at a professional level
- Ability to write clearly and concisely about technical issues and design reasoning, since feedback forms the primary deliverable
- Comfort identifying ambiguous or unclear task instructions and flagging those to researchers rather than guessing
- Familiarity with version and format differences across tools, and willingness to test across multiple platforms
- Availability for part-time project work on flexible schedules, typically 15-30 hours per week
Main skills
What the interview asks about
1.Formula robustness in spreadsheets
AI models often produce formulas that work on the author's data but fail with different ranges, empty cells, or new rows added by collaborators.
For example: “An AI wrote a sales dashboard with VLOOKUP formulas that work perfectly on today's data. If the source table gets sorted by a colleague tomorrow, what breaks?”
2.Document formatting across versions
AI frequently generates Word documents with complex styles and formatting that don't survive conversion to PDF or opening in older software versions.
For example: “You receive a beautifully formatted proposal from an AI system in Word 2024 format. A client using Office 2019 reports that tables overflow and formatting shifts. Why?”
3.Presentation slide logic
AI can produce slides with correct content but poor visual hierarchy, chart mislabeling, or logical gaps that derail boardroom presentations.
For example: “A model generates a quarterly review deck with three charts that individually look correct but tell a contradictory story about revenue trends. How do you frame this for the research team?”
4.Practical constraints versus solutions
Real professionals work under deadlines, legacy systems, and tool limitations that AI sometimes ignores, creating unusable outputs.
For example: “AI suggests using advanced pivot table features to solve an analyst's problem, but the company uses an older Excel version that lacks those functions. What's your assessment?”
5.File handoff and collaboration
AI-generated outputs often assume single-author workflows and break when multiple people edit, version-control, or convert formats.
For example: “An AI creates a complex budget template, but when three team members open it simultaneously in Excel Online, formulas recalculate incorrectly. What's failing?”
A task you may get
Design a task where AI must create a forecast spreadsheet with formulas. Evaluate two solutions: one using SUM functions, one using SUMIFS with named ranges. Test by adding data and converting to CSV. Document which approach works in production.
How to prepare
- Write down three complex tasks you've completed recently in office software (a budget, a report, a presentation) noting the tools, constraints, and quality standards.
- Review the structure of your most complex spreadsheet: identify dependencies, formulas, and areas where editing by someone else could introduce errors.
- Prepare two examples of AI-generated office documents you've seen (or can imagine), and think through where they'd fail under real-world use.
- Practice articulating technical issues clearly without jargon-your audience will be AI researchers, not office power users.
The facts
- Pay
- $60–70/hr
- Hours
- Part time, 40 hours a week
- Where
- Remote
- Field
- Miscellaneous
- Role type
- Talent network
- Posted
- 9/15/2026
We wrote this page from the public Mercor listing. It may be out of date, so read the full posting before you apply.